Installing MindSpore in Ascend by Source Code

• Installing MindSpore in Ascend by Source Code
  • Environment Preparation-automatic recommended
  • Environment Preparation-manual
    • Installing Python
    • Installing Ascend AI processor software package
    • Installing wheel and setuptools
    • Installing GCC
    • Installing git tclsh patch NUMA and Flex
    • Installing git-lfs
    • Installing CMake
  • Downloading the Source Code from the Code Repository
  • Compiling MindSpore
  • Installing MindSpore
  • Configuring Environment Variables
  • Installation Verification
  • Version Update

This document describes how to install MindSpore by compiling source code on Linux in an Ascend environment.

• If you want to configure an environment that can compile MindSpore on EulerOS 2.8 with Ascend AI processor software package installed, you may use an automatic installation script for one-click configuration, see Environment Preparation -automatic, recommended section. The script will automatically install the dependencies required to compile MindSpore.

• If your system is Ubuntu 18.04, CentOS 7.6, openEuler 20.03, or KylinV10 SP1, or some dependencies, such as Python and GCC, have been installed, it is advised to manually install MindSpore by referring to Environment Preparation-manual section.

Environment Preparation-automatic recommended

Before running the automatic installation script, you need to make sure that the Ascend AI processor software package is correctly installed on your system. If it is not installed, please refer to the section Installing Ascend AI processor software package to install it.

Run the following command to obtain and run the automatic installation script. The environment configured by the automatic installation script only supports compiling MindSpore>=1.6.0.

wget https://gitee.com/mindspore/mindspore/raw/master/scripts/install/euleros-ascend-source.sh
# install Python 3.7 by default
# the default value of LOCAL_ASCEND is /usr/local/Ascend
bash -i ./euleros-ascend-source.sh
# to specify the Python 3.9 installation and install the optional dependencies Open MPI
# and set LOCAL_ASCEND to /home/xxx/Ascend, use the following manners
# LOCAL_ASCEND=/home/xxx/Ascend PYTHON_VERSION=3.9 OPENMPI=on bash -i ./euleros-ascend-source.sh

This script performs the following operations:

• Install the compilation dependencies required by MindSpore, such as GCC, CMake, etc.
• Install Python3 and pip3 and set them as default.
• Install Open MPI if OPENMPI is set to on.

After the automatic installation script is executed, you need to reopen the terminal window to make the environment variables take effect.

The automatic installation script creates a virtual environment named mindspore_pyXX for MindSpore. Where XX is the Python version, such as Python 3.7, the virtual environment name is mindspore_py37. Run the following command to show all virtual environments.

conda env list

To activate the virtual environment, take Python 3.7 as an example, execute the following command.

conda activate mindspore_py37

Now you can jump to the Downloading the Source Code from the Code Repository section to downloading and compiling MindSpore.

For more usage, see the script header description.

Environment Preparation-manual

The following table lists the system environment and third-party dependencies required for building and installing MindSpore.

The following describes how to install the third-party dependencies.

Installing Python

Python can be installed by Conda.

Install Miniconda:

cd /tmp
curl -O https://mirrors.tuna.tsinghua.edu.cn/anaconda/miniconda/Miniconda3-py37_4.10.3-Linux-$(arch).sh
bash Miniconda3-py37_4.10.3-Linux-$(arch).sh -b
cd -
. ~/miniconda3/etc/profile.d/conda.sh
conda init bash

After the installation is complete, you can set up Tsinghua source acceleration download for Conda, and see here.

Create a virtual environment, taking Python 3.7.5 as an example:

conda create -n mindspore_py37 python=3.7.5 -y
conda activate mindspore_py37

Run the following command to check the Python version.

python --version

Installing Ascend AI processor software package

Ascend software package provides two distributions, commercial edition and community edition:

• Commercial edition needs approval from Ascend to download, for detailed installation guide, please refer to Ascend Training Solution 23.0.RC3.

• Community edition has no restrictions, choose 7.0.RC1.beta1 in CANN community edition, then choose relevant driver and firmware packages in firmware and driver. Please refer to the abovementioned commercial edition installation guide to choose which packages are to be installed and how to install them.

The default installation path of the installation package is /usr/local/Ascend. Ensure that the current user has the right to access the installation path /usr/local/Ascend of Ascend AI processor software package. If not, the root user needs to add the current user to the user group where /usr/local/Ascend is located.

Install the .whl packages provided in Ascend AI processor software package. If the .whl packages have been installed before, you should uninstall the .whl packages by running the following command.

pip uninstall te topi hccl -y

Run the following command to install the .whl packages in the default path. If the installation path is not the default path, you need to replace the path in the command with the installation path.

pip install sympy
pip install /usr/local/Ascend/ascend-toolkit/latest/lib64/te-*-py3-none-any.whl
pip install /usr/local/Ascend/ascend-toolkit/latest/lib64/hccl-*-py3-none-any.whl

The LD_LIBRARY_PATH environment variable does not work when the installation package exists in the default path. The default path priority is: /usr/local/Ascend/nnae is higher than /usr/local/Ascend/ascend-toolkit. The reason is that MindSpore uses DT_RPATH to support startup without environment variables, reducing user settings. DT_RPATH has a higher priority than the LD_LIBRARY_PATH environment variable.

Installing wheel and setuptools

After installing Python, use the following command to install it.

pip install wheel
pip install -U setuptools

Installing GCC

• On Ubuntu 18.04, run the following commands to install.

  sudo apt-get install gcc-7 -y

• On CentOS 7, run the following commands to install.

  sudo yum install centos-release-scl
  sudo yum install devtoolset-7

  After installation, run the following commands to switch to GCC 7.

  scl enable devtoolset-7 bash

• On EulerOS and openEuler, run the following commands to install.

  sudo yum install gcc -y

Installing git tclsh patch NUMA and Flex

• On Ubuntu 18.04, run the following commands to install.

  sudo apt-get install git tcl patch libnuma-dev flex -y

• On CentOS 7, EulerOS and openEuler, run the following commands to install.

  sudo yum install git tcl patch numactl-devel flex -y

Installing git-lfs

• On Ubuntu, run the following commands to install.

  curl -s https://packagecloud.io/install/repositories/github/git-lfs/script.deb.sh | sudo bash
  sudo apt-get install git-lfs -y
  git lfs install

• On CentOS 7, run the following commands to install.

  curl -s https://packagecloud.io/install/repositories/github/git-lfs/script.rpm.sh | sudo bash
  sudo yum install git-lfs -y
  git lfs install

• On EulerOS and openEuler, run the following commands to install.

  Select the appropriate version to download according to the system architecture.

  # x64
  curl -OL https://github.com/git-lfs/git-lfs/releases/download/v3.1.2/git-lfs-linux-amd64-v3.1.2.tar.gz
  # arm64
  curl -OL https://github.com/git-lfs/git-lfs/releases/download/v3.1.2/git-lfs-linux-arm64-v3.1.2.tar.gz

  Unzip and install.

  mkdir git-lfs
  tar xf git-lfs-linux-*-v3.1.2.tar.gz -C git-lfs
  cd git-lfs
  sudo bash install.sh

Installing CMake

• On Ubuntu 18.04, run the following commands to install CMake.

  wget -O - https://apt.kitware.com/keys/kitware-archive-latest.asc 2>/dev/null | sudo apt-key add -
  sudo apt-add-repository "deb https://apt.kitware.com/ubuntu/ $(lsb_release -cs) main"
  sudo apt-get install cmake -y

• Other Linux systems can be installed with the following commands.

  Choose different download links based on the system architecture.

  # x86 run
  curl -O https://cmake.org/files/v3.19/cmake-3.19.8-Linux-x86_64.sh
  # aarch64 run
  curl -O https://cmake.org/files/v3.19/cmake-3.19.8-Linux-aarch64.sh

  Run the script to install CMake, which is installed in the /usr/local by default.

  sudo mkdir /usr/local/cmake-3.19.8
  sudo bash cmake-3.19.8-Linux-*.sh --prefix=/usr/local/cmake-3.19.8 --exclude-subdir

  Finally, add CMake to the PATH environment variable. Run the following commands if it is installed in the default path, other installation paths need to be modified accordingly.

  echo -e "export PATH=/usr/local/cmake-3.19.8/bin:\$PATH" >> ~/.bashrc
  source ~/.bashrc

Downloading the Source Code from the Code Repository

git clone https://gitee.com/mindspore/mindspore.git

Compiling MindSpore

Go to the root directory of mindspore, then run the build script.

cd mindspore
bash build.sh -e ascend -S on

Where:

• In the build.sh script, the default number of compilation threads is 8. If the compiler performance is poor, compilation errors may occur. You can add -j{Number of threads} in to script to reduce the number of threads, for example, bash build.sh -e ascend -j4.
• By default, the dependent source code is downloaded from gitHub. When -S is set to on, the source code is downloaded from the corresponding gitee image.
• For details about how to use build.sh, see the script header description.

Installing MindSpore

pip install output/mindspore-*.whl -i https://pypi.tuna.tsinghua.edu.cn/simple

When the network is connected, dependencies of MindSpore are automatically downloaded during the .whl package installation. (For details about the dependency, see required_package in setup.py.) In other cases, you need to install it by yourself. When running models, you need to install additional dependencies based on requirements.txt specified for different models in ModelZoo. For details about common dependencies, see requirements.txt.

Configuring Environment Variables

If Ascend AI processor software is installed in a non-default path, after MindSpore is installed, export Runtime-related environment variables. /usr/local/Ascend in the following command LOCAL_ASCEND=/usr/local/Ascend denotes the installation path of the software package, and you need to replace it as the actual installation path of the software package.

# control log level. 0-DEBUG, 1-INFO, 2-WARNING, 3-ERROR, 4-CRITICAL, default level is WARNING.
export GLOG_v=2

# Conda environmental options
LOCAL_ASCEND=/usr/local/Ascend # the root directory of run package

# lib libraries that the run package depends on
export LD_LIBRARY_PATH=${LOCAL_ASCEND}/ascend-toolkit/latest/lib64:${LOCAL_ASCEND}/driver/lib64:${LOCAL_ASCEND}/ascend-toolkit/latest/opp/built-in/op_impl/ai_core/tbe/op_tiling:${LD_LIBRARY_PATH}

# Environment variables that must be configured
## TBE operator implementation tool path
export TBE_IMPL_PATH=${LOCAL_ASCEND}/ascend-toolkit/latest/opp/built-in/op_impl/ai_core/tbe
## OPP path
export ASCEND_OPP_PATH=${LOCAL_ASCEND}/ascend-toolkit/latest/opp
## AICPU path
export ASCEND_AICPU_PATH=${ASCEND_OPP_PATH}/..
## TBE operator compilation tool path
export PATH=${LOCAL_ASCEND}/ascend-toolkit/latest/compiler/ccec_compiler/bin/:${PATH}
## Python library that TBE implementation depends on
export PYTHONPATH=${TBE_IMPL_PATH}:${PYTHONPATH}

Installation Verification

i:

python -c "import mindspore;mindspore.set_context(device_target='Ascend');mindspore.run_check()"

The outputs should be the same as:

MindSpore version: __version__
The result of multiplication calculation is correct, MindSpore has been installed on platform [Ascend] successfully!

It means MindSpore has been installed successfully.

ii:

import numpy as np
import mindspore as ms
import mindspore.ops as ops

ms.set_context(device_target="Ascend")
x = ms.Tensor(np.ones([1,3,3,4]).astype(np.float32))
y = ms.Tensor(np.ones([1,3,3,4]).astype(np.float32))
print(ops.add(x, y))

The outputs should be the same as:

[[[[2. 2. 2. 2.]
   [2. 2. 2. 2.]
   [2. 2. 2. 2.]]

  [[2. 2. 2. 2.]
   [2. 2. 2. 2.]
   [2. 2. 2. 2.]]

  [[2. 2. 2. 2.]
   [2. 2. 2. 2.]
   [2. 2. 2. 2.]]]]

It means MindSpore has been installed successfully.

Version Update

After successfully executing the compile script build.sh in the root path of the source code, find the whl package in path output, and use the following command to update your version.

When upgrading from MindSpore 1.x to MindSpore 2.x, you need to manually uninstall the old version first:

pip uninstall mindspore-ascend

Then install MindSpore 2.x:

pip install mindspore*.whl

When upgrading from MindSpore 2.x:

pip install --upgrade mindspore*.whl